Tesla K80 with WSL2

[Stevoo01]

Hello together,

Is there a way to get a Tesla K80 detected in WSL2 or at least set up enough to use it in a Docker container.
If so, I would be happy if someone could help me with the setup. The topics I have found here so far have unfortunately not yet led to success. I'm not getting any further at the moment.

I have installed with WSL 2 Ubuntu. And follow these instructions.
https://developer.nvidia.com/cuda-downloads?target_os=Linux&target_arch=x86_64&Distribution=WSL-Ubuntu&target_version=2.0&target_type=deb_local

I also tryed this one.
https://www.forecr.io/blogs/installation/nvidia-docker-installation-for-ubuntu-in-wsl-2

And this one.
https://medium.com/htc-research-engineering-blog/nvidia-docker-on-wsl2-f891dfe34ab

This will not work, because no other card has the same chipset like the K80.
https://www.reddit.com/r/bashonubuntuonwindows/comments/kgx83d/any_way_to_run_nvidia_tesla_card_on_wsl2/

In end up every time, that wsl ubuntu dont see any nvida card.

Thanks for your help.

[DonnyVerduijn]

It probably is because the driver model of the K80 is set to TCC, which is unsupported in WSL2. You can verify this by running nvidia-smi -i 0 -q in command prompt/powershell.

See:
https://www.reddit.com/r/VFIO/comments/p9zcwi/how_to_workaround_tcc_to_wddm_on_nvidia_tesla/
https://stackoverflow.com/questions/56865611/how-to-enable-wddm-on-nvidia-p100
https://forums.developer.nvidia.com/t/how-to-open-wddm-mode-supporting-graphics-acceleration-under-windows-for-p100/166994

[MikeRuf]

This worked for me:

https://www.reddit.com/r/pcmods/comments/nhfwh7/guide_using_an_nvidia_tesla_k80_datacenter_gpu/

However, I'm running into problems using it with pytorch. Basic functionality works, but when I try to use more advanced features, I get:"

cuDNN error: CUDNN_STATUS_NOT_INITIALIZED

I'm guessing I didn't set up the toolkit properly on the WSL side, and it's not a problem with the virtualized access to the GPU.